ChatGPT does not cap the number of messages in a chat, but each model has a context window, so once a conversation exceeds it the earliest messages stop being visible to the model. Very long chats also load slowly and can fail to open. The fix is to export the thread, summarise it, and start a fresh conversation.
Short answer
ChatGPT does not cap the number of messages in a chat, but each model has a context window, so once a conversation exceeds it the earliest messages stop being visible to the model. Very long chats also load slowly and can fail to open. The fix is to export the thread, summarise it, and start a fresh conversation.
There is no message counter that stops you. What actually happens is subtler and more annoying: the conversation keeps working while quietly getting worse.
What the context window does
Every model reads a fixed amount of text per turn, counted in tokens across your prompt, the conversation so far, and the reply. Once the thread exceeds that budget, the oldest turns fall out of view. The model does not tell you this. It simply stops knowing what you agreed in message four.
Symptoms in order of appearance:
It forgets constraints you set early on.
It reintroduces mistakes you already corrected.
Replies get slower and the page becomes sluggish to scroll.
The conversation takes many seconds to load, or fails to load at all.
Roughly how much is too much
A useful rule of thumb rather than a hard number: once a chat passes a few hundred exchanges, or you have pasted several long documents into it, you are near the point where the beginning is gone. Attached files and images consume the budget faster than plain text.
Doing something about it
Start a new chat, deliberately. Ask the current one for a handover: "Summarise our decisions, constraints, and open questions as a briefing for a fresh conversation." Paste that into a new chat. You lose the transcript, not the context.
Keep the transcript outside ChatGPT. This is the part people skip. Export your data, convert the thread into a Markdown or Word document, and keep it with the project. Then the long chat can be abandoned without losing anything, and you can search the history in your own editor far faster than scrolling the sidebar.
Use Projects for continuity. Long lived work belongs in a Project with instructions and files, so the important context is loaded fresh each time rather than reconstructed from a bloated thread.
When a chat will not open at all
Very large conversations sometimes hang on load. Try a different browser, disable extensions, and give it time on a good connection. If it never opens, the data export is the way to retrieve the content: the JSON contains the full thread even when the interface cannot render it. Converting that file gives you back a chat you could not otherwise read.
Frequently asked questions
Is there a maximum number of messages in one ChatGPT chat?
No fixed limit, but the context window means older messages stop influencing replies, and very long chats become slow or unopenable.
Does starting a new chat lose my context?
It loses the transcript. Ask for a handover summary first and paste it into the new chat, and you keep the working context.
Why can't I open my longest conversation?
Rendering a very large thread can time out in the browser. The data export still contains it, and converting the export makes it readable.
Do attachments count towards the limit?
Yes. File and image content consumes context, often much faster than typed text.